Re: FW: How to caliculate the IO Demand for IOPS and ThroughtPut

From: kyle Hailey <>
Date: Fri, 24 Jun 2011 23:03:33 -0700
Message-ID: <>

My 2 cents for now

For data collection

dba_hist_sysmetric_summary- gives max, min, averages, stddev, etc for every hour for last week in AWR by default.

V$SYSMETRIC_HISTORY - max, min, avg, stddev etc for every minute for last hour. You can also pull out stats for every 15 secs for last 3 minutes.

I'd expect I/O to be less on disk if you are not using direct I/O many reads will come from UNIX file cache.
Even with direct I/O many reads will be coming from SAN cache. Then again SANs will do read aheads that might never get used by Oracle.

Also, for RAID5, I'd only expect 3x the IOPs for writes ( read, write, write vs write ) only.

Oracle can do a lot more I/O than just data block and redo operations. Check the stats

physical read IO requests
physical read total IO requests

physical read bytes
physical read total byte

On Thu, Jun 23, 2011 at 12:48 PM, Karl Arao <> wrote:

> Hi Robin,
> Yes I agree, the average normalizes the spikes.. so it's also good to have
> a validation of the IO numbers you are getting from Oracle from the OS
> numbers.
> Also I use these statistical tools for my data points
> this one is from Craig Shallahamer gives you a really nice visual
> distribution of your data points, I modified it a bit so he got that
> uploaded on his site already..
> And.. another thing that you can do when you are investigating on a slow
> subsystem and drilling down on a subsecond detail is to trace the session
> and use the Mr. Tools.. particularly the Mr. Skew.. ;) you can do a lot of
> statistical drill downs on 10046 trace using that tool
> --
> Karl Arao

Received on Sat Jun 25 2011 - 01:03:33 CDT

Original text of this message